Re: deploying both NFS client and server on the same machine trigger hungtask

[Trond Myklebust <trondmy@hammerspace.com>]

On Wed, 2024-12-04 at 15:45 +0000, Jeff Layton via Bugspray Bot wrote:
> Jeff Layton writes via Kernel.org Bugzilla:
>
> Li Lingfeng's analysis looks basically correct to me, though I think
> the nfsd_mutex is mostly a red-herring here.
>
> The client holds the shrinker rwsem and is trying to write back data.
> The server is trying to unregister a shrinker as part of server
> shutdown and has almost certainly stopped responding to requests at
> that point. The client is using hard RPCs, so it's going to retry the
> writeback indefinitely while holding the shrinker mutex, which will
> block server shutdown.
>
> I don't see a great way to fix this right offhand, though I wonder if
> localio might help mitigate this problem.
>
> View:
> https://bugzilla.kernel.org/show_bug.cgi?id=219550#c4
> You can reply to this message to join the discussion.

localio won't help with the task of returning a delegation, as per the
stack trace that was shown:
https://lore.kernel.org/linux-nfs/20241203-b219550c0-abf5589a5df5@bugzilla.kernel.org/

If the server is no longer responding to RPC calls, then the client
should probably just let the delegreturn call time out and expire after
2 lease periods.

--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com